High-dimensional Probability and Statistics — различия между версиями

Материал из Wiki - Факультет компьютерных наук
Перейти к: навигация, поиск
 
(не показано 9 промежуточных версии этого же участника)
Строка 12: Строка 12:
  
 
= Seminar content =
 
= Seminar content =
 +
 +
=== Probability ===
  
 
* (17.01.24) Example 2.4 from [[#wainwright|[Wainwright]]], Lemma 2.2 from [[#blm|[BLM]]]
 
* (17.01.24) Example 2.4 from [[#wainwright|[Wainwright]]], Lemma 2.2 from [[#blm|[BLM]]]
 +
 +
* (24.01.24) Appendix A and Exercise 2.2 of the second chapter of [[#wainwright|[Wainwright]]], Section 2.5.1 from [[#vershynin|[Vershynin]]]
 +
 +
* (31.01.24) Section 2.1.3 and Example 2.12 from [[#wainwright|[Wainwright]]]
 +
 +
* (07.02.24) Section 2.3 from [[#wainwright|[Wainwright]]]
 +
 +
* (24.02.24) Herbst's argument (Proposition 3.2 from [[#wainwright|[Wainwright]]]), Sub-additivity of the entropy (Theorem 4.22 from [[#blm|[BLM]]]), logorithmic Sobolev inequality for Gaussian random variables (Theorem 5.5 from [[#blm|[BLM]]])
 +
 +
* (07.03.24) PAC-Bayesian inequality. (Lemma 2.1 from [[#zhivotovsky | [Zhivotovsky]]])
 +
 +
* (14.03.24) Dimension-free concentration of sample covariance matrix in the spectral norm (Theorem 1.2 of [[#zhivotovsky | [Zhivotovsky]]])
 +
 +
* (21.03.24) Theorem 1.2 of [[#zhivotovsky | [Zhivotovsky]]], Concentration of Lipshitz and separately convex function of bounded random variables (Theorem 6.10 from [[#blm|[BLM]]]), Concentration of the supremum of an empirical process (Section 3.4 of [[#wainwright|[Wainwright]]])
 +
 +
=== Statistics ===
 +
 +
* (10.04.24) Linear Regression. Bayesian information criterion. (Theorem 2.4 from [[#rigollet | [Rigollet]]])
 +
 +
* (17.04.24) Restricted isometry property and epsilon-incoherence. (Incoherence section, pp.59-62 of [[#rigollet | [Rigollet]]])
 +
 +
* (24.04.24) Incoherence of a random matrix with independent Rademacher entries (Incoherence section, pp.59-62 of [[#rigollet | [Rigollet]]]). Empirical risk minimization and Rademacher complexity (Sections 4.1-4.2 of [[#wainwright|[Wainwright]]]). Bounds on the Rademacher complexity of finite and finite-dimensional classes can be found in [[#paris | [Paris]]], Theorems 6.1 and 6.3.
 +
 +
* (15.05.24) VC-dimension, Sauer's lemma. (Section 4.3 of [[#wainwright|[Wainwright]]]).
 +
 +
* (22.05.24) Packing-covering duality. (Lemma 5.12 of [[#van_handel|[van Handel]]]). Uniform bound on the metric entropy via VC-dimension (Theorem 7.16 of [[#van_handel|[van Handel]]])
 +
 +
* (29.05.24) Uniform bound on the metric entropy via VC-dimension (Theorem 7.16 of [[#van_handel|[van Handel]]])
 +
 +
* (05.06.24) Offset Rademacher Complexity (some parts of [[#puchkin|[Puchkin]]])
  
 
= References =
 
= References =
Строка 26: Строка 58:
  
 
<span id="blm">[BLM]</span> [https://disk.yandex.ru/i/7GOknoh1HYGEJQ Boucheron et al. Concentration inequalities]
 
<span id="blm">[BLM]</span> [https://disk.yandex.ru/i/7GOknoh1HYGEJQ Boucheron et al. Concentration inequalities]
 +
 +
<span id="zhivotovsky">[Zhivotovsky]</span> [https://disk.yandex.ru/i/3yAKKiHc73ZSZQ Nikita Zhivotovskiy. Dimension-free bounds for sums of independent matrices and simple tensors via the variational principle. Electron. J. Probab. vol. 29 (2024), article no. 13, 1–28.]
 +
 +
<span id="rigollet">[Rigollet]</span> [https://disk.yandex.ru/i/GW7kFWmdsrfClA Philippe Rigollet and Jan-Christian H¨utter. High-Dimensional Statistics. Lecture Notes]
 +
 +
<span id="paris">[Paris]</span> [https://disk.yandex.ru/i/CeIEx0MW2hpbrw Quentin Paris. Statistical Learning Theory. Lecture Notes]
 +
 +
<span id="puchkin">[Puchkin]</span> [https://proceedings.mlr.press/v195/puchkin23a.html Nikita Puchkin, Nikita Zhivotovskiy. Exploring Local Norms in Exp-concave Statistical Learning. COLT 2023]

Текущая версия на 16:55, 6 июня 2024

Classes

Wednesdays 16:20–17:40, in room R307.

Teachers: Alexey Naumov, Quentin Paris

Teaching Assistant: Fedor Noskov

Lecture content

Seminar content

Probability

  • (31.01.24) Section 2.1.3 and Example 2.12 from [Wainwright]
  • (24.02.24) Herbst's argument (Proposition 3.2 from [Wainwright]), Sub-additivity of the entropy (Theorem 4.22 from [BLM]), logorithmic Sobolev inequality for Gaussian random variables (Theorem 5.5 from [BLM])
  • (14.03.24) Dimension-free concentration of sample covariance matrix in the spectral norm (Theorem 1.2 of [Zhivotovsky])
  • (21.03.24) Theorem 1.2 of [Zhivotovsky], Concentration of Lipshitz and separately convex function of bounded random variables (Theorem 6.10 from [BLM]), Concentration of the supremum of an empirical process (Section 3.4 of [Wainwright])

Statistics

  • (10.04.24) Linear Regression. Bayesian information criterion. (Theorem 2.4 from [Rigollet])
  • (17.04.24) Restricted isometry property and epsilon-incoherence. (Incoherence section, pp.59-62 of [Rigollet])
  • (24.04.24) Incoherence of a random matrix with independent Rademacher entries (Incoherence section, pp.59-62 of [Rigollet]). Empirical risk minimization and Rademacher complexity (Sections 4.1-4.2 of [Wainwright]). Bounds on the Rademacher complexity of finite and finite-dimensional classes can be found in [Paris], Theorems 6.1 and 6.3.
  • (15.05.24) VC-dimension, Sauer's lemma. (Section 4.3 of [Wainwright]).
  • (22.05.24) Packing-covering duality. (Lemma 5.12 of [van Handel]). Uniform bound on the metric entropy via VC-dimension (Theorem 7.16 of [van Handel])
  • (29.05.24) Uniform bound on the metric entropy via VC-dimension (Theorem 7.16 of [van Handel])
  • (05.06.24) Offset Rademacher Complexity (some parts of [Puchkin])

References

links are available via hse accounts

[van Handel] Ramon van Handel. Probability in High Dimensions, Lecture Notes

[Vershynin] R. Vershynin. High-Dimensional Probability

[Wainwright] M.J. Wainwright. High-Dimensional Statistics

[BLM] Boucheron et al. Concentration inequalities

[Zhivotovsky] Nikita Zhivotovskiy. Dimension-free bounds for sums of independent matrices and simple tensors via the variational principle. Electron. J. Probab. vol. 29 (2024), article no. 13, 1–28.

[Rigollet] Philippe Rigollet and Jan-Christian H¨utter. High-Dimensional Statistics. Lecture Notes

[Paris] Quentin Paris. Statistical Learning Theory. Lecture Notes

[Puchkin] Nikita Puchkin, Nikita Zhivotovskiy. Exploring Local Norms in Exp-concave Statistical Learning. COLT 2023